Method and system for producing intent-based responses in a messaging platform

Abstract

A system for intent-based action recommendations and/or fulfillment in a messaging platform, preferably including and/or interfacing with: a set of user interfaces; a set of models; and/or a messaging platform. A method for intent-based action recommendations and/or fulfillment in a messaging platform, preferably including any or all of: receiving a set of information associated with a request; producing and sending a set of intent options; receiving a selected intent; generating a message based on the selected intent and/or the set of information; and providing the message.

We claim: 
     
         1 . A method comprising:
 receiving a set of messages of a conversation thread, the set of messages comprising:
 a first message posted to the conversation thread by a first user; and 
 a second message posted to the conversation thread by a second user, wherein at least one of the first message and the second message comprises a query message; 
   receiving an action selection from at least one of the first user and the second user; at a dialog manager comprising a first set of trained models, based on the action selection, determining an action prompt for a large language model (LLM);   at the LLM:
 receiving the action prompt and the set of messages; and 
 in response to receiving the action prompt and the set of messages, generating a response message, the response message responsive to the query message and to the action prompt; 
   generating an edited message upon receiving edits to the response message; posting the edited message to the conversation thread; and   refining the first set of trained models with the response message and the edited message.   
     
     
         2 . The method of  claim 1 , further comprising, before receiving the action selection:
 at the first set of trained models, determining a plurality of action selection options, the plurality of action selection options comprising the action selection; and   providing the plurality of action selection options to the user.   
     
     
         3 . The method of  claim 1 , wherein the dialog manager comprises a set of guardrail models comprising a first stage model structured as a low-latency model and a second stage model structured as a high-latency model, wherein the first stage model is structured to pass classifications of the conversation thread to the second stage model for further evaluation. 
     
     
         4 . The method of  claim 3 , further comprising: in response to an output of the second stage model characterizing high-risk content of the conversation thread, executing a generative action comprising automatically generating a message and posting the message to the conversation thread. 
     
     
         5 . The method of  claim 1 , further comprising: receiving, at an action server comprising a second set of trained models, a context of the conversation thread and instructions comprising a persona prompt and a language in which responses for the conversation thread should be generated. 
     
     
         6 . The method of  claim 1 , further comprising, after posting the edited message:
 receiving a second query message, the second query message different from the query message;   at the first set of trained models, based on the second query message, determining a second action prompt for the LLM, the second action prompt different from the action prompt;   at the LLM:
 receiving the second action prompt and the second query message; and in response to receiving the second action prompt and the second query message, generating a second response message, the second response message responsive to the second query message and to the action prompt, wherein generating the second response message comprises: at a classifier model of the first set of trained models, based on at least the second query message, selecting a selected message from a corpus of predetermined messages, wherein the second response message is determined based on the selected message. 
   
     
     
         7 . The method of  claim 1 , wherein the action selection pertains to initiating a quiz. 
     
     
         8 . The method of  claim 1 , wherein the conversation thread is provided within a messaging platform of a digital mental health platform. 
     
     
         9 . The method of  claim 1 , wherein the first user is a therapy-providing entity. 
     
     
         10 . The method of  claim 1 , further comprising providing at least one of first user and the second user an option to schedule a therapy session, through the dialog manager. 
     
     
         11 . A method comprising:
 receiving a set of messages of a conversation thread, the set of messages comprising:
 a first message posted to the conversation thread by a first user; and 
 a second message posted to the conversation thread by a second user, wherein at least one of the first message and the second message comprises a query message; 
   receiving an action selection from at least one of the first user and the second user; at a first set of trained models, based on the action selection, determining an action prompt for a large language model (LLM);   at the LLM:
 receiving the action prompt and the set of messages; and in response to receiving the action prompt and the set of messages, generating a response message, the response message responsive to the query message and to the action prompt, wherein generating the response message comprises: at a classifier model of the first set of trained models, based on at least the query message, selecting a selected message from a corpus of predetermined messages, wherein the response message is determined based on the selected message; and 
   posting the response message to the conversation thread.   
     
     
         12 . The method of  claim 11 , wherein the conversation thread is provided within a messaging platform of a digital mental health platform. 
     
     
         13 . The method of  claim 11 , wherein the first user is a therapy-providing entity. 
     
     
         14 . The method of  claim 11 , wherein the dialog manager comprises a set of guardrail models comprising a first stage model structured as a low-latency model and a second stage model structured as a high-latency model, wherein the first stage model is structured to pass classifications of the conversation thread to the second stage model for further evaluation. 
     
     
         15 . The method of  claim 14 , wherein, upon receiving a high-risk classification of content of the set of messages from the set of guardrail models, the method further comprises scheduling a clinical therapy session through the dialog manager. 
     
     
         16 . The method of  claim 11 , further comprising: receiving, at an action server comprising a second set of trained models, a context of the conversation thread and instructions comprising a persona prompt according to which generated messages should be posted to the conversation thread. 
     
     
         17 . The method of  claim 11 , wherein determining the action prompt is performed at a generative model of the first set of trained models. 
     
     
         18 . The method of  claim 11 , wherein the set of messages further comprises at least one message posted to the conversation thread by a non-human agent. 
     
     
         19 . The method of  claim 11 , wherein the response message is posted to the conversation thread in near-real time in response to receiving the action selection and the set of messages. 
     
     
         20 . A system of a digital mental health platform, the system comprising:
 a messaging platform accessible through a set of user interfaces of a set of user devices;   a set of models stored in non-transitory computer-readable media and structured to perform a set of actions, the set of models comprising:
 a dialog manager comprising a first set of trained models and 
 a large language model (LLM), 
 wherein the set of actions comprises:
 at the dialog manager, receiving a set of messages of a conversation thread from the messaging platform, the set of messages comprising: 
 a first message posted to the conversation thread by a first user; and 
 a second message posted to the conversation thread by a second user, wherein at least one of the first message and the second message comprises a query message; 
 from the messaging platform, receiving an action selection from at least one of the first user and the second user; 
 at a dialog manager, determining an action prompt for the large language model (LLM) based on the action selection; 
 at the LLM, receiving the action prompt and the set of messages; and 
 in response to receiving the action prompt and the set of messages, generating a response message, the response message responsive to the query message and to the action prompt; 
 generating an edited message upon receiving edits to the response message within the messaging platform; and 
 posting the edited message to the conversation thread.